Image segmentation using circularly spread MR images

Abstract

Describes a new segmentation method using an artificial neural network for brain magnetic resonance (MR) images. In the proposed method, the authors spread an MR image circularly to recognize several regions in the brain. In the spread image, the cerebral regions including white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF) align in layers from bottom to top. The above regions are segment by the following procedures: First, the contour of the whole cerebral region including CSF is detected from the spread image by using a Gaussian filter and a differential operator. The region outside the contour is then eliminated. Cerebral regions WM, GM, and CSF are eliminated by an artificial neural network which has three layers. The inputs of the neural network are a pixel value and its normalized vertical and horizontal locations in the spread image. The proposed method was examined using the T2-weighted images, and it was able to segment WM, GM, and CSF regions accurately.